Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and...

22
Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University Shinchon-dong, Sudaemoon-ku, Seoul 120-749, Korea [madoka, sbcho]@candy.yonsei.ac.kr PRICAI- 2000

Transcript of Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and...

Page 1: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design

Hee-Su Kim and Sung-Bae Cho

Computer Science Department, Yonsei UniversityShinchon-dong, Sudaemoon-ku, Seoul 120-749, Korea

[madoka, sbcho]@candy.yonsei.ac.kr

PRICAI-2000

Page 2: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

2

Agenda

Motivation

Backgrounds

System development

Knowledge-based encoding

Experimental results

Conclusion and future works

Page 3: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

3

Before the Industrial Revolution : Customers have few choices on buying their clothes

After the Industrial Revolution :Customers can make their choices with very large variety

Near Future :Customers can order and get clothes of their favorite design

Manufacturer

Oriented

ConsumerOriented

Changes on Consumer Economy

Motivation

Page 4: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

4

• Almost all consumers are non-professional at design

• To make designers contact all consumers is not effective

• Need for the design system that can be used by non-professionals

Need for Interaction-based System

Motivation

Page 5: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

5

Fashion design

– To make a choice within various styles that clothes can take

Three shape part of fashion design

– Silhouette

– Detail

– Trimming

Backgrounds

Fashion Design

Page 6: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

6

t=0;

Initialize Population

Evaluate P(t);

while not done do

t=t+1;

P’=Select Parents P(t)

Recombine P’(t);

Mutate P’(t);

Evaluate P’(t);

P=Survive P, P’(t);

end while

Crossover Mutation

Genetic Algorithm

Backgrounds

Page 7: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

7

Crossover / Mutation Fitness Evaluation

Reproduction

Initial Population

Population

User Selection

GA IGA

Fitness Function

Interactive Genetic Algorithm

Backgrounds

Page 8: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

8

Related Works

Virtuosi System (Nottingham Trent University, 1998) AutoCAD with ApparelCAD plug-in (Autodesk co.)

– Fashion design aid system for professionals only Manual Evolutionary Design Aid System (Nakanishi, 1996)

– Often produces impractical designs

Backgrounds

Interactive GA KB Encoding

Apply evolutionary Computation using domain specific knowledge

Page 9: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

9

Overview

System Development

Decode

User Fitness

Combine

Display

Interactive Genetic Algorithm

OpenGL Program

GA operation Reproduce

Models ofeach part

Page 10: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

10

VRML : Simply get 3D but too slow

OpenGL : Faster but not easy to implement

Use GLUT library with OpenGL

– Reduce the burden of programming OpenGL

3D Modeling Method

System Development

Page 11: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

11

Modeling by 3D Studio MAX

System Development

Page 12: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

12

IGA Fashion Design Aid System

System Development

Page 13: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

13

Knowledge-based Encoding

Gene Encoding

Search space size=34*8*11*8*9*8=1,880,064

A : Neck and body style(34) E : Skirt and waistline style(9)

C : Arm and sleeve style(11)

B : Color(8)

D : Color(8)

……

A B C D E F

Total 23 bits

F : Color(8)

Page 14: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

14

Example Design from a Genotype

Knowledge-based Encoding

001010 101 0101 011111 0111

High Green WhiteTrumpetPurpleMelon

Page 15: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

15

Schema Theorem

The instances of schema H in particular generation t+1, m(H, t+1), can be expressed in terms of m(H, t)

Schemata with short defining length, low order, above-average fitness receive exponentially increasing trials in subsequent generations

Knowledge-based Encoding

mc

Homc

pHol

Hp

f

HftHmtHm

pl

Hp

f

HftHmtHm

)(1

)(1

)(),()1,(

11

)(1

)(),()1,( )(

Page 16: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

16

Experimental Environment

Subjects – 10 male and female student, no background on

fashion design

Crossover rate : 0.5 (1-point crossover)

Mutation rate : 0.05 (Binary mutation)

10 generations with elitist preserving

Request for each subjects– Find out most cool-looking design with given system

Experimental Results

Page 17: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

17

Convergence Test for Cool-Looking Design

Experimental Results

0

10

20

30

40

50

60

70

80

1 2 3 4 5 6 7 8 9 10

Generation

Fitn

ess

Valu

e

Average Fitness Value Best Fitness Value

Page 18: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

18

Subjective Test

Experimental Results

Examples of searched design which gives cool feeling

Page 19: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

19

Fitness Changes for each Encoding Method

Experimental Results

0

10

20

30

40

50

60

70

1 2 3 4 5 6 7 8 9 10

Generation

Ave

rage

Fitn

ess

Knowledge- based Encoding

Sequential Encoding

Page 20: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

20

Relative Satisfaction for each Encoding Method

Experimental Results

Page 21: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

21

Example Solution Design and Frequency of Each Solution Schema

- 10123456789

10

0 1 2 3 4 5 6 7 8 9 10

Generations

Freq

uenc

y

010100***************** (Slit body design)******011************** (White body color)*********1011********** (Sleeveless arm design)****************0100*** (Scooter skirt design)********************100 (Blue skirt color)

Experimental Results

Page 22: Genetic Algorithm with Knowledge-based Encoding for Interactive Fashion Design Hee-Su Kim and Sung-Bae Cho Computer Science Department, Yonsei University.

22

Conclusion and Future Works

Knowledge-based Encoding in Interactive Genetic Algorithm for a Fashion Design Aid System– Based on Knowledge of fashion design– Compared with sequential encoding by several experiments

Future Works– Adding up extra design elements such as textile : To enlarge the se

arch space– Clustering : To avoid Genetic drift caused by small population size– Direct Manipulation : To accelerate convergence with relatively short

generation